Remote Sensing Image Quality Assessment Based On Statistical Distribution

In the process of imaging,transmission and storage of remote sensing images,limited to the imperfection of imaging system,coding algorithm and storage device,and the interference of cloud and fog radiation during imaging,resulting in remote sensing image suffers different degree of distortion and degradation,which brings great difficulties to the remote sensing image processing and application.Therefore,design a reasonable remote sensing image quality assessment method,it is not only applicable to the evaluation of remote sensing image quality in remote sensing image real-time processing system,but also can be used as feedback for the optimization of remote sensing imaging system and image processing algorithm,which has important research significance and application value.In this thesis,based on the assumption that degradation will change the change the statistical distribution of remote sensing images,through the systematic study of existing theories and methods,the remote sensing image quality assessment method is studied deeply.The main research contents are summarized as follows:A remote sensing image quality assessment method based on regional contrast combined with gradient similarity is proposed,and a remote sensing image distortion database is constructed.By analyzing the human visual brightness characteristics,the regional contrast map of the reference remote sensing image is firstly calculated,then using gradient to characterized of structural degradation of remote sensing image,and the gradient similarity map were computed between the gradients of reference and distorted remoting sensing images,finally,combined the region contrast maps and gradient similarity map and obtain the objective quality score of remote sensing image.Experimental results on the constructed remote sensing image distortion database shows the proposed method has a high consistency with human subjective quality scores.A no-reference remote sensing image quality assessment method based on texture features statistical distribution is proposed.The proposed method predicts the distortion of the remote sensing image by analyzing the change of the texture information before and after the distortion.By extract the gray level co-occurrence matrix of the remote sensing image to represent the global texture change of the image,the local binary patterns of gradient map to represent the local texture change of the image.Thereafter,a support vector regression is applied to predict the mapping function between the texture features and the human subjective perception,constructs the remote sensing image quality prediction model.Experimental results reveal the rationality and the validity of the proposed method.A blind remote sensing image quality assessment without subjective score training is proposed.The proposed method does not need the subjective score to training.By extract the undistorted reference remote sensing image training set's spatial domain and frequency domain statistical features,and fitted it to the multivariate Gaussian mixture model,to constructs the remote sensing image's standard feature statistics distribution.Then the distorted remote sensing image quality score is acquired by introducing a simple difference metric between the standard statistics distribution and the distorted remote sensing image statistics distribution.The proposed method does not need the subjective perception quality score,experimental results show that the proposed method is more versatility and robust.